Merging Expert
Merging expert models, a technique increasingly used in machine learning, aims to improve efficiency and performance by combining the knowledge of multiple specialized models. Current research focuses on developing effective merging strategies within architectures like Mixture-of-Experts (MoE), often incorporating techniques like gating networks to selectively combine expert outputs or leveraging usage frequency to prioritize the merging of highly relevant experts. This approach holds significant promise for enhancing the capabilities of large language models and other complex systems by reducing computational costs while maintaining or improving accuracy across diverse tasks, such as multimodal perception and misinformation detection.
Papers
Channel Merging: Preserving Specialization for Merged Experts
Mingyang Zhang, Jing Liu, Ganggui Ding, Xinyi Yu, Linlin Ou, Bohan Zhuang
A Survey on Inference Optimization Techniques for Mixture of Experts Models
Jiacheng Liu, Peng Tang, Wenfeng Wang, Yuhang Ren, Xiaofeng Hou, Pheng-Ann Heng, Minyi Guo, Chao Li